radiometric correction of digital uas multispectral

Radiometric Correction of Digital UAS Multispectral Imagery Using Free

and Open Satellite Surface Reflectance Images

^^ ^

^ ^C U S L

1

*Saket Gowravaram, *Haiyang Chao, #Andrew L. Molthan,

*Nathaniel Brunsell, and +Tiebiao Zhao

* The University of Kansas

#NASA Short-term Prediction Research and Transition Center (SPoRT)

+XPeng Motors

(E): [email protected]

Jan. 14, 2020, American Meteorological Society Annual Meeting.

Nov. 16, 2012

Aviation Safety Research and Development

January 14, 2020

Cooperative Unmanned Systems Laboratory^

^ ^^ ^

2

Outline

I. Introduction

II. Comparison of Remote Sensing Data at Different Spatial

Resolutions (Landsat 8 OLI, Sentinel 2 MSI, and NEON

Phase One)

III. Satellite-based Empirical Radiometric Inter-Calibration

IV. Validation of identified Radiometric Inter-Calibration

functions using high-resolution NEON data

V. Conclusions & Future Work

Nov. 16, 2012


January 14, 2020


^ ^^ ^

Motivations

• Accurate and low-cost high-resolution multispectral field observations are critical to

applications including precision agriculture and disaster damage assessment (e.g., hail,

tornados, fire, etc.);

• Spatiotemporal enrichment of existing free and open satellite remote sensing data at low

spatial resolution (Landsat 8 OLI, Sentinel 2 MSI) using high resolution uncalibrated UAS

data.

Contributions

• Proposed novel methods for radiometric correction of high resolution digital UAS

multispectral orthomosaic maps using free and open satellite surface reflectance images;

• Generated high-resolution UAS images of calibrated reflectance and derived products (e.g.,

NDVI) using the identified mapping functions to facilitate the examination of higher

resolution features and inferences which are crucial for applications including vegetation

monitoring and disaster damage tracking;

• Validation of the generated UAS calibrated SR images using high-resolution (1 m/pix)

NEON surface reflectance images acquired by manned aircraft over a 32 subplot hay field

(10 × 10 m).

Introduction

3

Nov. 16, 2012


January 14, 2020


^ ^^ ^

4

KHawk UAS Platform

Sensing Payload: PeauPro82 RGB Camera and

modified multispectral (R,G,NIR) Camera

Bands FWHM

Wavelength

(nm)

Peak

Wavelength

(nm)

OLI Blue 452.02 –

512.06

482.04

Green 532.74 –

590.07

561.41

Red 635.85 –

673.32

654.59

NIR 850.54 –

878.79

864.67

GoPro Hero 4

True Color

Blue 398.7 – 505.5 458.5

Green 475.9 – 602.9 532.9

Red 583.9 – 710 627.6

GoPro Hero 4

Modified

Multispectral

NIR 825.4 – 880 852.7

The NIR band from the modified multispectral camera and the R,G, and B bands from the unmodified

true color camera are used to prevent noise in the visible bands due to NIR leakage.

~- ~-'t,,,.,, ..._H~ L ..._H~

Nov. 16, 2012


January 14, 2020


^ ^^ ^

Workflow

5

Ortho UAS

Multispectral Maps

in DN (GSD: 0.1

m/pix.)

Calibrated SR

Airborne

Multispectral

Images

(GSD: 1 m/pix.)

Calibrated SR

Satellite Multispectral

Images

(GSD: 30 m/pix.)

Downsampled Ortho

UAS Multispectral

Maps in DN (GSD:

30 m/pix.)

Identification of Mapping

Function between DN and SR

Multispectral Images Using

Selected Regions

32 Subplots of

Hay Field

High-resolution

Calibrated

Orthorectified SR UAS

Multispectral Maps

(GSD: 1 m/pix.)

ValidationInter-Calibration

·-··-··-··-··-··-··-··1 ,----------• • I .

l ,_- ' I I I I

.

L----t---- .. ' __J\

I • • -··-··-··-··-··-··-·· -

Nov. 16, 2012


January 14, 2020


^ ^^ ^

6

Outline

I. Introduction

II. Comparison of Remote Sensing Data at Different

Spatial Resolutions (Landsat 8 OLI, Sentinel 2 MSI,

and NEON Phase One)





Nov. 16, 2012


January 14, 2020


^ ^^ ^

Description KHawk UAS

GoPro

NEON Phase

One

Sentinel 2

MSI

Landsat 8

OLI

Platform Type UAS Aircraft Satellite Satellite

Revisit Times < 1 day 1 year 5 days 16 days

Bands 4 426 13 9

Spatial

Resolution

0.1 m/pix 1 m/pix 10 m/pix 30 m/pix

Remote Sensing System Description

7

KHawk UAS NEON Aircraft Landsat 8

https://www.usgs.gov/land-resources/nli/landsat/landsat-8?qt-

science_support_page_related_con=0#qt-science_support_page_related_con

https://www.neonscience.org/data-collection/airborne-remote-sensing

https://www.usgs.gov/land-resources/nli/landsat/landsat-8?qt-science_support_page_related_con=0#qt-science_support_page_related_con

https://www.neonscience.org/data-collection/airborne-remote-sensing

Nov. 16, 2012


January 14, 2020


^ ^^ ^

Landsat 8 OLI, Sentinel 2A MSI, and

NEON Comparison

8/38

Region used for Comparison

The Sentinel 2A MSI Image (10 m/pix) and the Neon NDVI image (1 m/pix.) are

sampled later to 30 m/pix by pixel aggregation for comparison.

Landsat 8 OLI NDVI (30 m/pix) S2A MSI NDVI (10 m/pix) Neon NDVI (1 m/pix)

Platform NIR Red NDVI

S2A-L8 0.9207 0.9313 0.9417

NEON-L8 0.8972 0.9091 0.9207

2D Correlation

Nov. 16, 2012


January 14, 2020


^ ^^ ^

Landsat 8 OLI, Sentinel 2A MSI, and

NEON Comparison

9/38

The reflectance values of multispectral imagery remain about the same at

different spatial resolutions (30 m to 1 m) for regions with vegetation

homogeneity.

The red line represents “y = x”.

Nov. 16, 2012


January 14, 2020


^ ^^ ^

10

Outline

I. Introduction



Phase One)

III. Satellite-based Empirical Radiometric Inter-

Calibration




Nov. 16, 2012


January 14, 2020


^ ^^ ^

Uncalibrated UAS DN Maps and Calibrated L8

SR Images

11/38

KHawk UAS RGB DN Map

(GSD: 0.1 m/pix)KHawk UAS NIR DN Map

(GSD: 0.1 m/pix)

L8 OLI NIR SR Map

(GSD: 30 m/pix)

• The orthorectified KHawk DN maps, RGB (L), NIR (M), and calibrated L8 OLI NIR SR (R)

maps are shown below.

Nov. 16, 2012


January 14, 2020


^ ^^ ^

Satellite-based Empirical Radiometric Inter-

Calibration (SERI)

12/38

The SERI method involves three steps:

• Pixel-aggregation of the UAS DN maps to match the spatial resolution of the reference image

(L8 SR);

• Extract regions from the scene that represent the range of the reflectance values of each

respective spectral band;

– The extracted regions should also exhibit low vegetation variability in smaller scales:

• Identification of the correction function between the UAS DN mean values and the L8 SR

values corresponding to the extracted regions.

Extracted Regions (black circles) shown on the L8 NDVI image

High-resolution view of

one of the extracted

regions

Coefficient of Variation of small-scale

features for each regions

Nov. 16, 2012


January 14, 2020


^ ^^ ^

13/38

• u ....

.... ,. . ... • • Oll1 • • • _ ... _ .... • •

l !- • .. ~ " » •u • I • . "" 1- • ~u !!I..,.

~•U&M.utta•.,,.. Cl.QI ...,.--·-1\t,A II.I .......... MIR•o.nt 0.0, R"-o.-

RM8E • IUl1-

0 0 .. • .. ,. uo 1• , .. UIO aoo 2IO .. '° ,. ,. ,..., 1'0 IIO --NRl!lllo(WM) ICH----(0-255)

... om ... OJII , • ~

O#T • • - • _o.ot ~- ~ ~

s ... "' • "' 0.0. i ! • • o...,. !I o.oa • • 5

Cl.OIi • ...... ~•UODlt Nu,o l\a•-G.olOl•-N11,1 R"•1.m1 0A

.... _.. o.ot -•Ma ,.

• ---0.0, 0.0, .. .. 100 11:o , .. 180 1IO .. .. 100 120 1'0 UIO IIO

---Dllt{0.:1:1¥) IOi"""-91,,. DNo (GaS)

Nov. 16, 2012


January 14, 2020


^ ^^ ^

14

Outline

I. Introduction



Phase One)


IV. Validation of identified Radiometric Inter-

Calibration functions using high-resolution NEON

data


Nov. 16, 2012


January 14, 2020


^ ^^ ^

Validation using Selected Hay Field

15/38

• At the start of our studies in 2000, the field was dominated by introduced

C3 hay-grasses, Bromus inermis (Smooth Brome) and Schedonorus

arundinaceus (Tall Fescue).

https://foster.ku.edu/long-term-studies-secondary-succession-and-community-

assembly-prairie-forest-ecotone-eastern-kansas

https://foster.ku.edu/long-term-studies-secondary-succession-and-community-assembly-prairie-forest-ecotone-eastern-kansas

Nov. 16, 2012


January 14, 2020


^ ^^ ^

High-Resolution KHawk NDVI Map (1 m/pix)

16/38

• Validation at selected research hay field (32 subplots with each 10 × 10 meter)

– 4 subplots each row and 8 subplots each column.

Nov. 16, 2012


January 14, 2020


^ ^^ ^

High-Resolution NEON NDVI Map (1 m/pix)

17/38

• Ground truth from NEON data for validation of calibrated UAS map at

high resolution (1 m).

Nov. 16, 2012


January 14, 2020


^ ^^ ^

Analysis Using 32 sub-plot Hay Field

18/38

NEON NDVI at 1 m KHawk NDVI at 1 m

• The average reflectance and NDVI within each selected subplot are

compared between UAS and NEON data.

Nov. 16, 2012


January 14, 2020


^ ^^ ^

19/38

Band 𝑅2 RMSE

NIR 0.7237 0.0153

Visible (RGB

Combined)

0.8165 0.0071

Analysis Using 32 sub-plot Hay Field• The NIR band and visible spectrum reflectance values predicted by the KHawk UAS map are

compared with the measured reflectance values observed in the NEON image.

Nov. 16, 2012


January 14, 2020


^ ^^ ^

20

Outline

I. Introduction



Phase One)


IV. Validation of identified Radiometric Inter-

Calibration functions using high-resolution NEON

data


Nov. 16, 2012


January 14, 2020


^ ^^ ^

Conclusions

• For the ROI the spectral values at different resolutions: 1, 10, and 30 m/pix demonstrated a high

correlation of ~ 90 % indicating that the vegetation showed low variability, therefore, validating the

idea of using low-resolution L8 SR images to inter-calibrate the downsampled UAS DN maps;

• The developed SERI for DN-SR calibration showed low RMSE of 0.033 for the NIR band and

0.011, 0.0005, and 0,0059 for the R,G, and B bands respectively;

• Finally, the calibrated KHawk SR images were compared to high-resolution airborne NEON

images to validated the proposed inter-calibration technique;

• A 32 sub-plot Hay field was used to perform this comparison which should very good correlation of

𝑅2 = 0.724 for the NIR band and 𝑅2 = 0.817 for the visible bands;

Future Work

• Test our proposed algorithm on different data sets and different ROI;

• Conduct flight tests for NDVI-time series analysis and perform comprehensive analysis with

corresponding L8 OLI images over different days and times.

Conclusions & Future Work

21


January 14, 2020


^ ^^ ^

Thank You!

22

radiometric correction of digital uas multispectral

Documents